OK, here we go (# time /usr/pkg/bin/openssl speed )

==> openssl_speed-NO_CFLAGS.txt <==
compiler: gcc -DOPENSSL_SYSNAME_ULTRASPARC -fPIC -DOPENSSL_PIC -DDSO_DLFCN
-DHAVE_DLFCN_H -DOPENSSL_NO_KRB5 -DOPENSSL_NO_IDEA -DOPENSSL_NO_MDC2
-DOPENSSL_NO_RC5 -O2 -DTERMIOS -DB_ENDIAN -DMD32_REG_T=int -O2 -Wall
-DMD5_ASM

                  sign    verify    sign/s verify/s
rsa  512 bits 0.007187s 0.000699s    139.1   1431.3
rsa 1024 bits 0.041468s 0.002119s     24.1    472.0
rsa 2048 bits 0.256691s 0.007283s      3.9    137.3
rsa 4096 bits 1.729724s 0.026498s      0.6     37.7
                  sign    verify    sign/s verify/s
dsa  512 bits 0.006405s 0.007834s    156.1    127.7
dsa 1024 bits 0.020348s 0.025166s     49.1     39.7
dsa 2048 bits 0.071391s 0.087142s     14.0     11.5
  369.00s real   368.50s user     0.19s system

==> openssl_speed-CFLAGS.txt <==
compiler: gcc -DOPENSSL_SYSNAME_ULTRASPARC -fPIC -DOPENSSL_PIC -DDSO_DLFCN
-DHAVE_DLFCN_H -DOPENSSL_NO_KRB5 -DOPENSSL_NO_IDEA -DOPENSSL_NO_MDC2
-DOPENSSL_NO_RC5 -O2 -pipe -mcpu=ultrasparc -mtune=ultrasparc -m64 -mvis
-DTERMIOS -DB_ENDIAN -DMD32_REG_T=int -O2 -Wall -DMD5_ASM

                  sign    verify    sign/s verify/s
rsa  512 bits 0.007418s 0.000681s    134.8   1468.5
rsa 1024 bits 0.040334s 0.002055s     24.8    486.7
rsa 2048 bits 0.247915s 0.007043s      4.0    142.0
rsa 4096 bits 1.670201s 0.025462s      0.6     39.3
                  sign    verify    sign/s verify/s
dsa  512 bits 0.006257s 0.007697s    159.8    129.9
dsa 1024 bits 0.019685s 0.025104s     50.8     39.8
dsa 2048 bits 0.068205s 0.084632s     14.7     11.8
  368.60s real   368.07s user     0.08s system

the only real gain I see is less time in 'system' while running.
else, no real gain.

I plan to upgrade to -current and try using GCC4 to compare latter on.
Anyway, any remarks on those current results ?

TIA,
     Jo

On Fri, July 7, 2006 7:34 pm, Martin Husemann wrote:
> On Fri, Jul 07, 2006 at 05:49:32PM +0200, Joel CARNAT wrote:
>
>> what surprises me the most is the various values I get.
>
> I'd say that ssh-keygen is not cpu bound in your case.
> "openssl speed" might be a better test.
>
>
> Martin
>
>